Variational Regression for Multi-Target Energy Disaggregation.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Energy Systems

Background:

• Non-intrusive load monitoring (NILM) systems using deep learning offer accurate end-use detection.
• Current NILM models often employ a suboptimal 'one vs. one' strategy, training separate models for each appliance.
• This leads to high computational costs for training and inference due to numerous parameters and models.

Purpose of the Study:

• To propose a novel, efficient NILM system capable of disaggregating multiple appliances using a single model.
• To reduce the computational burden associated with training and deploying individual appliance models.

Main Methods:

• Development of a multi-appliance power disaggregation model based on a multi-target regression neural network.
• The architecture features a variational encoder with convolutional layers and shared parameters across multiple regression heads.
• The model simultaneously estimates the individual power consumption of all target appliances from the total installation load.

Main Results:

• The proposed multi-target regression network effectively disaggregates power consumption for multiple appliances concurrently.
• Experimental comparisons demonstrate the model's performance against existing state-of-the-art multi- and single-target NILM approaches.
• The single-model approach shows promise in reducing the overall complexity and cost of NILM systems.

Conclusions:

• The novel multi-appliance NILM model offers a more efficient alternative to single-appliance strategies.
• This approach significantly reduces training and inference costs while maintaining high accuracy.
• The developed system advances the practical deployment of NILM technology for energy management.